Latest AI News

OpenAI Robotics Lead Caitlin Kalinowski Resigns, Citing Surveillance and Lethal Autonomy Concerns

Grammarly's "Expert Review" feature, launched in August 2025, generates AI writing suggestions supposedly inspired by named subject matter experts, including The Verge's editor-in-chief Nilay Patel, editors David Pierce, Sean Hollister, and Tom Warren, plus dozens of other journalists and professors, none of whom gave permission. Parent company Superhuman defended the practice, saying experts appear because their published works are publicly available. Critics note the feature contains inaccurate job titles, broken source links, and misleading comment-style formatting that mimics real editorial feedback.

Roblox is rolling out AI-powered real-time chat rephrasing to replace blocked profanity with acceptable alternatives instead of showing strings of #####. Authored by VP Rajiv Bhatia and Chief Safety Officer Matt Kaufman, the feature targets in-experience text chat between age-verified users. For example, "Hurry tf up!" becomes "Hurry up!" Roblox also upgraded its text filters with large reasoning models, reducing false negatives for personal information sharing by 20x. The long-term goal is retiring the ##### system entirely.

Anthropic Launches Local Scheduled Tasks in Claude Code Desktop

Microsoft announced it will continue offering Anthropic's Claude AI models to customers through platforms including Microsoft 365, GitHub Copilot, and Azure AI Foundry, despite the Pentagon labeling Anthropic a supply chain security risk. Microsoft's lawyers concluded the designation does not prevent use outside the Department of War. Microsoft is the first major company to publicly commit to keeping Anthropic products available. Anthropic plans to challenge the Pentagon designation in court. Microsoft has a $5 billion investment in Anthropic and a broader $30 billion Azure cloud commitment.

Anthropic partnered with Mozilla to use Claude Opus 4.6 to find security vulnerabilities in Firefox. Over two weeks, Claude discovered 22 vulnerabilities, 14 of which Mozilla classified as high-severity, representing nearly a fifth of all high-severity Firefox vulnerabilities remediated in 2025. Claude scanned nearly 6,000 C++ files and submitted 112 unique reports. Most issues were patched in Firefox 148.0. Anthropic also tested exploit development, spending $4,000 in API credits, but Claude only successfully built working exploits in two cases.

OpenAI has launched Codex Security, formerly known as Aardvark, as an application security agent now in research preview for ChatGPT Pro, Enterprise, Business, and Edu customers with free usage for one month. The tool uses frontier models and agentic reasoning to detect and patch complex vulnerabilities, reducing false positives by over 50% and over-reported severity by 90% during beta. It has already scanned 1.2 million commits, found 792 critical issues, and reported CVEs to OpenSSH, GnuTLS, and other major open-source projects.

Claude launches Claude Marketplace to simplify enterprise AI tool procurement in limited preview

Perplexity Introduces Skills for Perplexity Computer, Reusable Actions It Learns Once

OpenAI launched ChatGPT for Excel in beta, an add-in powered by GPT-5.4 that lets users build, update, and analyze spreadsheet models using plain language. GPT-5.4 Thinking scored 87.3% on OpenAI's internal investment banking benchmark, up from 43.7% with GPT-5. New financial data integrations with Moody's, Dow Jones Factiva, MSCI, Third Bridge, and MT Newswire are also live, with FactSet coming soon. The Excel add-in is available to Business, Enterprise, Pro, and Plus users in the U.S., Canada, and Australia.

Anthropic CEO Dario Amodei confirmed the company received a letter from the Department of War on March 4 designating it a supply chain risk to national security, and Anthropic plans to challenge the designation in court. Amodei clarified the ruling narrowly applies only to Claude's use within specific Department of War contracts, not all customers. He also apologized for a leaked internal post, noting it was written hastily and does not reflect his considered views.

Anthropic published a labor market study introducing a new metric called observed exposure, which combines theoretical LLM capability with real-world Claude usage data to measure AI displacement risk. The top exposed occupations include Computer Programmers at 75% coverage, Customer Service Representatives, and Data Entry Keyers at 67%. Highly exposed workers tend to be older, female, more educated, and earn 47% more on average. BLS projections show weaker job growth for exposed roles, and early data suggests hiring of younger workers has slowed in those occupations.

Meta is facing a U.S. class action lawsuit over privacy violations tied to its Ray-Ban AI smart glasses, filed by plaintiffs Gina Bartone and Mateo Canu and represented by Clarkson Law Firm. An investigation by Swedish newspapers revealed that subcontractors at a Kenya-based firm reviewed user footage, including nudity and sexual content. Meta marketed the glasses with slogans like "designed for privacy, controlled by you," but users cannot opt out of the data review pipeline. Over seven million people bought the glasses in 2025.

Luma AI has launched Luma Agents, a creative AI agent platform powered by its new Unified Intelligence model family, starting with Uni-1, a multimodal model trained on audio, video, image, language, and spatial reasoning. The agents coordinate with external models including Google's Veo 3, ByteDance's Seedream, and ElevenLabs to handle end-to-end creative work across text, images, video, and audio. Early customers include Publicis Groupe, Adidas, and Mazda. Luma Agents is now available via API.

Revised: Netflix has acquired InterPositive, an AI filmmaking tools startup founded by Academy Award-winning director Ben Affleck in 2022. The company builds filmmaker-focused AI tools trained on proprietary datasets filmed on a controlled soundstage to capture real production workflows and cinematic logic. Affleck will join Netflix as Senior Advisor. Netflix Chief Product Officer Elizabeth Stone said the deal reflects a shared belief that innovation should em

OpenAI research finds that current reasoning models struggle to control their chain-of-thought reasoning, which is good news for AI safety. Tested across 13 frontier models including GPT-5 Thinking, OpenAI o3, Claude Sonnet 4, and DeepSeek R1, controllability scores ranged from just 0.1% to 15.4%. Controllability increases with model size but drops with longer reasoning and more reinforcement learning training. OpenAI plans to report CoT controllability alongside monitorability in future system cards, starting with GPT-5.

Google, Meta, Microsoft, Oracle, OpenAI, Amazon, and xAI signed a White House ratepayer protection pledge on March 4, 2026, committing to cover electricity costs and grid upgrade expenses tied to their data centers. President Trump hosted the roundtable to address bipartisan concerns about rising household electricity bills, which climbed 13 percent nationally in 2025. Companies must negotiate agreements with utilities and state governments, and pay for new infrastructure even if data centers don't use all generated power.

The UK's Information Commissioner's Office (ICO) is writing to Meta after a Swedish newspaper investigation by Svenska Dagbladet and Goteborgs-Posten revealed that workers at Nairobi-based subcontractor Sama were reviewing sensitive videos captured by Ray-Ban Meta smart glasses, including footage of people undressing and watching pornography. Meta confirmed contractors sometimes review shared content to improve user experience, but critics say users are unaware their footage is seen by humans despite disclosures buried in Meta's privacy policies.

Nvidia CEO Jensen Huang said at the Morgan Stanley Tech, Media and Telecom conference that his company's investments in OpenAI and Anthropic will likely be its last, citing anticipated IPOs as closing the investment window. However, analysts note other factors: Nvidia's $100 billion OpenAI pledge shrank to $30 billion in the final deal, Anthropic CEO Dario Amodei publicly criticized Nvidia's chip sales to China, and the Pentagon recently blacklisted Anthropic. Observers suggest Nvidia may be exiting a situation that has grown increasingly complicated.

AWS has launched OpenClaw on Amazon Lightsail, making it easier to run the open-source autonomous private AI agent on AWS infrastructure. Lightsail instances come pre-configured with Amazon Bedrock as the default AI model provider, requiring no additional setup to start chatting. OpenClaw can connect to messaging apps like WhatsApp, Discord, and Telegram to manage emails, browse the web, and organize files. A 4 GB memory plan is recommended, and pricing follows Lightsail's on-demand hourly rate plus Bedrock's per-token model.

Microsoft has released Phi-reasoning-vision-15B, a 15 billion parameter open-weight multimodal reasoning model available on Microsoft Foundry, HuggingFace, and GitHub. Using a mid-fusion architecture with a SigLIP-2 Naflex dynamic-resolution vision encoder built on the Phi-Reasoning backbone, the model was trained on just 200 billion multimodal tokens. It achieves roughly 75% accuracy across benchmarks including MathVista, MMMU, ChartQA, and ScreenSpot-v2, outperforming similarly fast models while requiring ten times less compute than larger rivals like Qwen and Kimi-VL.

Junyang Lin, a central technical leader on Alibaba's Qwen AI team, announced he was stepping down on X just one day after Alibaba launched its Qwen 3.5 Small Model series, which includes four open-weight multimodal models spanning 0.6B to 9B parameters. The circumstances remain unclear, and Alibaba did not respond to comment requests. Colleagues described his departure as an immense loss, with one teammate writing that leaving was not Lin's choice. Elon Musk had praised the new Qwen 3.5 models for their impressive intelligence density.

Google has updated NotebookLM with Cinematic Video Overviews, a new feature powered by Gemini 3, Nano Banana Pro, and Veo 3. Unlike the previous narrated slides format, the tool generates fluid animations and rich visuals tailored to user sources. Gemini acts as a creative director, making hundreds of structural and stylistic decisions, refining its own output for consistency. The feature launches in English today for Google AI Ultra subscribers aged 18 and older, available on web and mobile.

Google has rolled out Canvas in AI Mode to all U.S. users in English, adding new support for creative writing and coding tasks directly within Search. Users can access Canvas via the tool menu in AI Mode and describe what they want to build, receiving a working prototype in a side panel that draws on live web data and Google's Knowledge Graph. One early tester example included a scholarship-tracking dashboard. Users can view underlying code and refine outputs through conversational follow-ups.

OpenAI has launched the Learning Outcomes Measurement Suite, a framework developed with Estonia's University of Tartu and Stanford's SCALE Initiative to assess AI's long-term impact on learning. The suite goes beyond test scores by tracking engagement, metacognition, task persistence, and creativity over time. It follows a randomized study of 300 college students using ChatGPT's study mode, which showed roughly 15% higher microeconomics exam scores versus a no-AI control group. Validation is underway with nearly 20,000 Estonian students aged 16-18.

OpenAI has published a preprint showing that single-minus graviton tree amplitudes, long assumed to vanish, are in fact nonzero under special kinematic conditions called the half-collinear regime. GPT-5.2 Pro played a key role by deriving the gravitational extension from a prior gluon result using the directed matrix-tree theorem and producing a preliminary paper draft. The work, authored by researchers from Harvard, Cambridge, Vanderbilt, IAS, and OpenAI, also connects these amplitudes to an infinite-dimensional symmetry originally studied by Roger Penrose.

OpenAI Codex App Launches on Windows With Native Agent Sandbox and PowerShell Support

LTX Studio Launches Dubbing and Captions for Enterprise Video Localization

Anthropic submitted a proposal to compete in a $100 million Pentagon prize challenge to develop voice-controlled, autonomous drone swarm technology, according to people familiar with the matter. The submission came during tense negotiations between Anthropic and the Defense Department over the AI safety company's red lines for military use of its technology. The situation has since escalated: Defense Secretary Pete Hegseth ordered the Pentagon to bar its contractors and partners from any commercial activity with Anthropic.

Anthropic has updated its skill-creator tool for Claude with no-code testing and benchmarking features aimed at non-engineer skill authors. The update adds eval writing, benchmark mode tracking pass rates and token usage, parallel multi-agent eval runs, and A/B comparator agents to judge skill versions blindly. A new description optimizer helped improve triggering accuracy on 5 of 6 public document-creation skills. Updates are live on Claude.ai, Cowork, and as a Claude Code plugin on GitHub.

OpenAI CEO Sam Altman announced the company will amend its Defense Department deal to explicitly ban domestic mass surveillance of Americans, citing Fourth Amendment protections and FISA. Altman published an internal memo on X, admitting the deal was rushed on February 27 and looked opportunistic. He also stated intelligence agencies like the NSA cannot use OpenAI services without a contract modification, and that he would rather go to jail than follow an unconstitutional order.

U.S. uninstalls of OpenAI's ChatGPT mobile app surged 295% day-over-day on February 28 after news broke of its deal with the Department of Defense, compared to a typical 9% daily uninstall rate, according to Sensor Tower. One-star reviews also spiked 775% that day. Meanwhile, Anthropic's Claude saw U.S. downloads jump 51% after the company declined the same Pentagon deal over concerns about AI surveillance and autonomous weapons. Claude hit No. 1 on the U.S. App Store and, per Appfigures, its daily downloads surpassed ChatGPT's for the first time.

An investigation by Svenska Dagbladet and Göteborgs-Posten reveals that Meta Ray-Ban smart glasses are sending user data to Meta servers, where Kenyan workers at subcontractor Sama review intimate footage including nudity, sex scenes, and bathroom visits captured accidentally by wearers. Workers in Nairobi describe seeing bank cards, pornography, and private conversations. Despite retailer claims of full user control, network analysis showed the glasses require constant Meta server contact, and Meta's own terms confirm human review of AI interactions is mandatory.

Apple announced 14- and inch MacBook Pro models featuring the new M5 Pro and M5 Max chips, built on Apple's Fusion Architecture with Neural Accelerators in each GPU core. The chips deliver up to 4x faster AI performance than M4 Pro and M4 Max, and up to 8x versus M1 models. SSD speeds reach up to 14.5GB/s, double the previous generation. Starting storage increases to 1TB for M5 Pro and 2TB for M5 Max. Pre-orders open March 4, with availability March 11.

Deveillance Unveils Spectre I Smart Device to Block Unwanted Audio Recordings

xAI has released Grok 4.20 Beta 2, an updated iteration of its Grok AI model. The release highlights two key improvements: enhanced instruction following, which means the model better adheres to user prompts and specified constraints, and reduced hallucinations, indicating greater factual accuracy in its responses. The beta label suggests the update is still undergoing refinement before a full release. No additional benchmarks, technical specifications, or broader deployment timelines were shared alongside the announcement.

Anthropic has begun rolling out a Voice Mode feature in Claude Code, its AI-powered coding assistant, initially making it available to just 5% of users. The company plans to gradually expand access to a larger portion of its user base over the coming weeks. Voice Mode allows developers to interact with the coding tool verbally rather than typing, which could meaningfully streamline workflows and reduce friction during software development tasks, offering a more natural interface for programmers.

Claude adds Memory to free plan with easier import and export options

Qualcomm announced the Snapdragon Wear Elite, a new 3nm chip targeting AI wearables like pins, pendants, and display-free smart glasses rather than replacing its existing W5 Plus smartwatch chip. The Elite features both an eNPU for low-power tasks and a Hexagon NPU capable of handling two billion parameters on-device at up to 10 tokens per second. It also delivers 40 percent better GPS power efficiency, 9V quick charging for 50 percent battery in 10 minutes, and supports 5G, Bluetooth 6.0, and Linux.

Anthropic's Claude is surging in popularity after the company refused to let the Department of Defense use its AI for mass domestic surveillance or autonomous weapons. President Trump retaliated by ordering federal agencies to stop using Anthropic products, while OpenAI struck its own Pentagon deal. The backlash drove Claude to the top of Apple's U.S. App Store, with Anthropic reporting free users up over 60% since January and paid subscribers more than doubling. TechCrunch outlines how to export ChatGPT data and import it into Claude.

The U.S. Supreme Court declined on March 2 to hear a case by Missouri computer scientist Stephen Thaler over whether AI-generated art can receive copyright protection. Thaler applied in 2018 for a copyright on A Recent Entrance to Paradise, created by his AI system DABUS, but the U.S. Copyright Office rejected it in 2022, ruling creative works must have human authors. Lower courts upheld that decision, and the Trump administration urged the court not to intervene, leaving the human-authorship requirement intact in U.S. copyright law.

At MWC 2026 in Barcelona, HONOR unveiled its Robot Phone, Magic V6 foldable, and first humanoid robot. The Robot Phone features a 4DoF gimbal, 200MP sensor, and embodied AI with emotional body language. Magic V6 is 8.75mm thin with a 6,660mAh silicon-carbon battery at 25% silicon content, dual LTPO 2.0 screens up to 6,000 nits, and Snapdragon 8 Elite Gen 5. The humanoid robot targets shopping, workplace, and companionship scenarios. HONOR also debuted MagicPad 4 with a 3K OLED display and MagicBook Pro 14 with Intel Core Ultra Series 3.

Anthropic has launched an import feature for Claude that lets users transfer their preferences and context from other AI providers in under a minute. The two-step process requires users to copy a provided prompt into any rival AI chatbot to extract saved context, then paste the results into Claude's memory settings. Claude then updates its memory so the first conversation feels familiar rather than starting from scratch. The feature is available on the free plan and keeps project context separate to prevent bleed-over between different tasks.

Apple has unveiled the new iPad Air powered by the M4 chip, featuring 12GB of unified memory, 120GB/s memory bandwidth, and a core Neural Engine. The tablet is up to 30 percent faster than the M3 model and 2.3x faster than M1. New Apple-designed chips N1 and C1X enable Wi-Fi 7 and faster cellular. It runs iPadOS 26 with a new Liquid Glass design. Prices start at $599 for the inch and $799 for the inch, with preorders March 4 and availability March 11.

Apple announced the iPhone 17e on March 2, 2026, as the most affordable member of the iPhone 17 family, starting at $599 with 256GB of storage. It features the A19 chip, Apple's C1X cellular modem that is 2x faster than the C1 in iPhone 16e, a 48MP Fusion camera with optical-quality 2x Telephoto, and a 6.inch Super Retina XDR display with Ceramic Shield 2. MagSafe support enables 15W wireless charging. Pre-orders begin March 4, with availability March 11.

Treasury to End Use of Anthropic Products, Including Claude Platform, Under Trump Order

Alibaba Qwen Launches Qwen3.5 Small Model Series (0.8B to 9B)